Alvin Lang
Feb 05, 2026 18:35
Claude Opus 4.6 scores 23% higher on financial analysis benchmarks, adds Excel and PowerPoint integrations for investment banking workflows.
Anthropic dropped Claude Opus 4.6 on February 5, positioning its latest AI model as a direct play for financial services workflows. The headline number: a 23 percentage point improvement over Claude Sonnet 4.5 on the company’s internal Real-World Finance benchmark, which tests roughly 50 investment and financial analysis use cases.
The model now scores 60.7% on Vals AI’s Finance Agent benchmark—a 5.47% jump from Opus 4.5—which evaluates performance on SEC filing research. It also hits 76% on TaxEval, another external benchmark testing tax-related reasoning.
Where Analysts Actually Work
The real story here isn’t just benchmark scores. Anthropic is pushing Claude directly into the tools finance professionals use daily: Excel and PowerPoint.
Claude in Excel now handles pivot tables, chart modifications, conditional formatting, and what Anthropic calls “finance-grade formatting.” The integration supports multi-file drag-and-drop and auto-compaction for long conversations—addressing the copy-paste hell that plagues anyone building complex financial models across multiple tabs.
Claude in PowerPoint launches in beta for Max, Team, and Enterprise users. The AI reads existing layouts, fonts, and master slides before generating new content, theoretically letting analysts build client-ready decks without starting from scratch.
The Productivity Claim
Anthropic’s marketing materials show side-by-side comparisons of commercial due diligence outputs—the kind of acquisition analysis work they say “would typically take a senior analyst two to three weeks to complete.” First-pass quality has improved noticeably, according to partners already testing the system.
“Creating financial PowerPoints that used to take hours now takes minutes,” said Aabhas Sharma, CTO at Hebbia. Nico Christie, co-founder of Shortcut AI, called it “a watershed moment for spreadsheet agents.”
Lloyd Hilton, Head of Hg Catalyst, noted the model handles “unstructured data and intelligently working with minimal prompting to meaningfully automate complex analysis.”
What’s New Under the Hood
Opus 4.6 ships with a 1-million-token context window, letting it process massive datasets in single sessions. The model also improved on BrowseComp and DeepSearchQA benchmarks, which test information extraction from large, unstructured document sets—critical for anyone doing earnings call analysis or regulatory filing reviews.
Cowork, Anthropic’s desktop app feature, now lets finance teams kick off multiple analyses simultaneously while steering Claude’s approach on each deliverable. A corporate finance plugin provides pre-built workflows for journal entries, variance analyses, and reconciliation.
The Fine Print
Anthropic isn’t claiming full autonomy. “Users should continue to review Claude’s outputs to ensure it meets their specifications; particularly for high-stakes work, human judgment remains essential,” the company noted in its release.
For crypto and fintech firms evaluating AI integration, Opus 4.6 represents the clearest signal yet that foundation model companies are moving beyond chatbot interfaces toward embedded enterprise tools. The question now: how quickly will competing models from OpenAI and Google match these finance-specific capabilities?
Claude Opus 4.6 is available now on all paid Claude plans. The PowerPoint integration remains in research preview for higher-tier subscribers.
Image source: Shutterstock
Credit: Source link
